ylqi / gl-rg Goto Github PK

The CIDER metric file is determined. Isn't there a mismatch between index of metric and data of the random Ground_Truth is selected? And I plan to do more experiments. Could you please post the refFile for calculating CIDEr? Thanks

data文件缺少train相关的文件，请问可以提供完整train的数据吗？非常感谢！！

颜教授您好，不好意思，想再请教一下 在跑 install.md当中 ./preprocess.sh的时候 出现如图片中所示问题 其实评价指标部分的代码是有的 但是它读取不到 出现 ModuleNotFoundError vscode 中 子目录读取不到父级目录中的代码显示 ModuleNotFoundError

（pycharm里面可以手动设置cider和coco-caption两个目录为资源根目录 但是不清楚 vscode里面怎么去做这种设置） 抱歉 应该蛮好解决的但是我实在是没找到怎么去设置

Thank you for your great work! There are different numbers of captions for every video in MSVD dataset, such as 29, 42……But I found that the shape of msvd_train_evalscores.pkl is [1200,17], why there are only 17 captions' scores for every video in training set?

I remember the the standard split contains 1.2K training videos, 100 validation videos, and 670 test videos, so why you did not use the other 200 test videos? Thank u!

Hi,
I notice you said "Our long-range encoder is pre-trained on the video-to-words dataset (k=300 words) extracted from MSR-VTT or MSVD" in your paper, I wanna to know whether the whole datasets(train, valid, test) were used in your pretraining phase. If so, I think it will lead to serious information leakage. Could you please release your pretraining code? thx : ).

颜教授您好，不好意思，想请教一下 论文中对应三个encoder特征抽取层面的代码好像在项目中没有找到，可以放出来嘛？想了解一下 传统方式抽取特征的话 代码实现层面是怎么操作的，谢谢

Hello, Thank you for sharing your amazing work. I have some questions:
1- Can you shortly explain in steps not as code how to obtain the metric scores m( ˆ S) of all ground truths captions
2- in file "GL-RG\data\preprocess\compute_scores.py" :-

    for i in range(args.seq_per_img):
        logger.info('taking caption: %d', i)
        preds_i = {v: [gt_refs[v][i]] for v in videos}

        # removing the refs at i
        if args.remove_in_ref:
            gt_refs_i = {v: gt_refs[v][:i] + gt_refs[v][i + 1:] for v in videos}
        else:
            gt_refs_i = gt_refs

        for scorer, method in scorers:
            score_i, scores_i = scorer.compute_score(gt_refs_i, preds_i)

Why both gt_refs_i and preds_i are equel to gt_refs , but I think preds_i should be the predicted caption based on the initial training with XE

Hi! Thank you for your amazing work! I want to ask how can I generate the msrvtt_train_evalscores.pkl? I see the shape is [6513,20], so which caption is respectively corresponding to the every 20 scores of a video?

Traceback (most recent call last):
  File "test.py", line 130, in <module>
    assert opt.feat_dims == test_loader.get_feat_dims()
AssertionError

I did not make any change before this.

您好，非常感谢您的工作！请问训练的主函数代码可以开源吗？train.py只是定义了一些方法而没有程序入口和主函数。